Fig. 4
From: Population-specific calibration and validation of an open-source bone age AI
Bootstrap re-partitioning (n = 1000) into calibration (⅓) and test (⅔) sets and its effect. Left: the distribution of sex-specific regression parameters (slope and intercept) obtained from the calibration sets and the resulting mean absolute difference (MAD) and root mean squared error (RMSE) in the test sets. Right: the resulting corrections (i.e. difference between Deeplasia and Deeplasia-GE) as median and bootstrapped 95% CI. The solid lines indicate the correction derived from the selected test set.
